Elg, Fredrik, "An Approach for Understanding Learning and Decision Making in Complex Dynamic Systems", 1996

An approach for understanding learning and decision making
in complex dynamic systems

by Fredrik Elg!

People often fail in controlling complex dynamic situations. Research on dynamic decision
making serves the purpose of learning about how people come to think, learn and act in
complex dynamic and opaque situations, where the general objective is to draw general
conclusions about the nature of tasks and differences between people in learning and decision
making. Dynamic decision making research differ from traditional decision making research
by explicitly addressing issues of feedback in the task. For a definition of dynamic decision
making see Brehmer (1992). Introducing concepts from system dynamics offer new
possibilities for research on dynamic decision making by presenting a framework for
understanding real life systems. System dynamics also offers a possibility to provide
transparency to complex microworlds, provides ideas on how to improve learning in and
about complex dynamic systems and, finally, system dynamics methodology can be used to
ease microworld construction and improve aspects of ecological validity. The full version of
this paper deal in more detail with research on dynamic decision making and issues on the
development of methods for understanding learning and decision making in and about
complex dynamic systems.

Dynamic decision making and system dynamics combined

“The importance of gaining an awareness of the enemy before the enemy gains a similar
awareness" (quoted in Gilson 1995) was expressed by Oswald Boelke during WW I and
represents a dynamic decision problem where failure can cause disaster. The concept of
dynamic decision making was originally described by Edwards (1962) and extended by
Rapoport (1975), where typical research in dynamic decision making concern problems of
learning about the correct assessment of complex dynamic and opaque task situations and the
implementation of measures to achieve some desired state of affairs (Brehmer 1992). Control
of any system can be understood within a control theory framework and can be described in
terms of the constraints of the system (Vicente and Rasmussen 1992 for a framework).
System dynamics and dynamic decision making have these underlying assumptions in
common. In accordance with Ashby's (1956) demand for requisite variety, a decision maker
needs to realise that controlling complex systems needs complex control and a correct
representation of the system.

The human mind is limited in its capacity and is characterized by essentially sequential
processing, limited memory, selective perception and reliance on cognitive simplification
mechanisms (Hogarth 1981). As a consequence of cognitive stress a decision maker adopts a
strategy to make it possible to exert some control over a task. The level of cognitive load is
adapted to fit to the cognitive resources at hand (Bainbridge 1979). In identifying problems
people conceptualize the current environment and draw conclusions according to the
perceived state of the situation. System dynamics provide a means of improving our
understanding of the systems we wish to control, and in a well understood control problem

1 Uppsala University, Department of Psychology, Box 1854, S-751 48 UPPSALA, e-mail:
fredrik.elg@psyk.uu.se, Tel +46 (0) 18-18 25 90, Fax +46 (0)18- 18 22 02

193
there is no cognitive stress and interaction is based on real-time, multi-variable, synchronous
co-ordination (Rasmussen and Pejtersen, 1992). The presence of a correct model of the
decision situation is the single most important component in being able to make a correct
decision (Brehmer, 1992). When the mental model of a situation and the actual situation
differ, errors come as an inevitable consequence. However, research about control of systems
face great problems in that the complexity and nature of most dynamic systems outside
laboratory conditions makes it impossible to obtain useful and valid information (Hoc 1989),
not to say expensive. By using a computer simulated microworlds, simulating essential
features of the real world (Brehmer and Dérner, 1993) it is possible to facilitate the testing of
scenarios and to control learning in the interaction with these microworlds in an experimental
way. In fact the microworld paradigm is the main source of theory and research on decision
making in dynamic systems (Brehmer 1992). Some research in dynamic decision making has
been performed within the system dynamics research tradition, where subjects have been
presented to specific decision contexts such as oil tanker shipping, real-estate (Bakken 1993),
and experimental markets (Kampmann 1992) etc. See Brehmer (1992), and Frensch & Funke
(1995), for overviews of previous research. Typical for most studies however is that the
microworld tasks are novel to the subjects and not based on problems in real life.

By combining disciplines, paradigms can be created, developed and explored (Kuhn 1970).
System dynamics have two general goals, the first is to understand complex dynamic systems
through analysis of these systems through modelling. This first objective is approached by
doing empirical and theoretical studies on real life problems and implementing the results of
these studies in a simulation model for analysis. The second goal is to improve peoples
systems reasoning skills and develop skill in understanding, conceptualizing and building
models of systems. System dynamics can provide specific and well built representations of
real life problems that easily can be turned into microworlds by adding an interface to the
system simulations, for instance by building management flight simulators (Sterman 1992).
Simulations can represent the structure and complexity of aggregate dynamic systems with
great fidelity and permit controlled manipulations of the decision contexts in the system
(Sterman 1989a and others). Using the skills developed in system dynamics it is thus possible
to represent specific rather than general characteristics of real life systems. By being able to
integrate and model the micro behavior of decision makers in aggregate structures it is
possible to model aggregate dynamic systems (Sterman 1989b). This however assumes that
we know and can simulate the behavior of decision makers.

"It appears that the experimental exploration of dynamic decision making strategies in
aggregate systems is feasible. The fidelity and flexibility of simulation models enables
the investigator to construct rich, complex decision making environments... the
marriage of experimental research of judgement with realistic simulation models thus
offers a reproducible procedure to explore the endogenous generation of macro
behavior from the micro structure of complex systems" (Sterman, 1989a, p 330)

Computational cognitive modelling is a growing field in psychology with research attempting
to make more or less dynamic models of the human decision making process (see Cacciabue
et al. 1992 and Démer and Wearing, 1995 for examples). A greater understanding of decision
makers cognitive structure and processes can contribute to improved learning methodologies
and the development of psychologically relevant decision support systems and modes of
control. One route to the development of more accurate measures of peoples learning and

194
decision making is by moving closer to real world decision making. A way of reaching this
goal is through the microworld paradigm. This is however not an easy path to walk and there
are serious problems with adopting microworlds that simulate naturalistic decision situations
(Brehmer and Dorner 1993), By combining system dynamics ‘and dynamic decision making
two general objectives can be achieved. First, microworlds are easier to make using the tools
and methods used in system dynamics and can thus flexibly be made to fit different structures
of tasks, different subjects and system demands. Second, what is learit from dynamic
decision making and the experimental paradigm can be integrated in the system dynamics
modelling program, to provide simulated aggregate behavior of decision makers and test the
intended rationality of decision rules in the simulation models.

The integration of research traditions can be achieved through the development of the
microworld paradigm. An obvious first step in the combination of the two methodologies is
to fit the management flight simulators of the system dynamics tradition with "flight
recorders". By providing measures of decision maker's performance and mental models this
can provide important information about how decision makers assess their situation, learn and
make decisions. In view of this and the great potential gains from the respective disciplines,
the marriage between system dynamics and dynamic decision making may provide new
means to understand the formulation and revision of decision makers mental representations
and decisions. However, understanding peoples mental models will still remain a difficult
methodological problem because of the aggregate nature of peoples mental models, and
especially concerning naturalistic decision problems.

Directions for future research

By providing microworlds with proximity to every day decisions "gut level'-responses may
be facilitated and validity may be improved in a range of experimental settings. Combined
with measures of the decision process and subjects mental models, system dynamics
modelling procedures may enable us to create flexible microworlds to explore learning and
decision making where critical ecological issues may be addressed and the validity of
experiments can be improved and explored. It may be possible to establish measures of
learning based on psychologically relevant frameworks (See Vicente and Rasmussen 1992 for
an example). Based on the learning history of decision makers, measures of mental processes
can be improved, providing opportunity to move closer to real life learning and decision
making in decision research. Typical problems to address may concern subjects mental
representations of systems in terms of situation assessment, and to model the consequences of
subjects mental representations. To what extent is this knowledge automated into schemas,
for example in the skill-rule-knowledge framework (Rasmussen and Pejtersen 1992). Other
issues to deal with are the importance of situation characteristics (surface characteristics) for
situation awareness, and how we can move toward system understanding and more
autonomous decision making, following the ideas of Piaget (1932). Another problem area
concern the representation of solutions to problems, see for example Reason 1990 and Klein
1993. Measures of general performance will remain a problem as these are limited by our
own understanding of the problems under study. Eventually we will be better able to answer
how systems are learned and represented in our minds and how this knowledge is transformed
into decisions, where new insights can help us design systems that are better able to facilitate
the formation of correct mental representations and efficiently communicate the
developments of a system. If not, at least we will be able to model the consequences of our

failure. 155
References

Ashby W. R. (1956) An introduction to cybernetics, London, Chapman and Hall

Bakken B. E. (1993) Learning and transfer of understanding in dynamic decision environments,
Dissertation, Massachusetts Institute of Technology

Bainbridge, L. (1979) Forgotten altematives in skill and workload, Ergonomics, 21, 169-185

Brehmer B. (1992), Dynamic decision making: Human control of complex systems, Acta
Psychologica, 81, 211-241, North-Holland

Brehmer B. and Démer D, (4993)-Experiments with computer-simulated microworlds: Escaping both
the narrow straits of the laboratory and the deep blue sea of the field study, Computers in
Human Behavior, 9, 171-184

Cacciabue, P.C., Kjaer-Hansen, J. & Reason, J. (1992) Cognitive modeling and human machine
interactions in dynamic decision environments, In MOHAWK, Models of human activities in
work context, Separate papers 1, Roskilde, Ris National Laboratory

Démer, D and Wearing. A. J. (1995) Complex problem solving: Toward a (computer simulated)
theory, In Frensch, P. A. and Funke J, Complex problem solving, The European perspective,
Lawrence Erlbaum Associates, New Jersey

Edwards W.(1962), Dynamic decision theory and probabilistic information processing. Human
Factors, 4, 59-73

Frensch P. A. and Funke J. (1995) Complex problem solving, The European perspective, Hillsdale,
NJ, Lawrence Erlbaum Associates

Gilson, R (1995) Human factors, 37 (1) 3-4

Hoc J. M. (1989), Strategies in controlling a continuos process with long response latencies: Needs
for computer support to diagnosis, International Journal of Man-Machine Studies, 30, 47-67

Hogarth R. M. (1981), Beyond discrete biases: Functional and dysfunctional aspects of judgmental
heuristics, Psychological Bulletin, 90, (2), 197-217

Kampmann, C. P. (1992) Feedback complexity and market adjustment: An experimental approach,
Dissertation, Massachusetts Institute of Technology

Klein G. A. (1993) A recognition primed decision, RPD, model of rapid decision making, in Klein G.
A., Orasanu, O., Calderwood, R. and Zsambok E. (eds.) Decision making in action: Models
and methods, Ablex Publishing Corp., 139-47,

Kuhn, T. S. (1970) Structure of scientific revolutions, 2ed., University of Chicago Press

Piaget J. 1932, The Moral judgement of the Child, Routledge and Kegan Paul, London

Rapoport A. (1975), Research paradigms for the study of dynamic decision behavior, In: D. Wendt
and C. Vlek (eds.), Utility, probability and human decision making. Dordrecht: Reidel

Rasmussen, J. and Pejtersen, A. M. (1992), MOHAWK taxonomy: Implications for design and
evaluation, in MOHAWC, Models of human activities in work context, 37- 55 , Roskilde, Ris
National Laboratory

Reason , J. (1990) Human Error, Cambridge University Press

Sterman J. D. (1989a) Misperception of feedback in dynamic decision making. Organizational
Behavior and Human Decision Processes, 43, 301-335

Sterman J. D. (1989b) Modeling managerial behavior: Misperceptions of feedback in a dynamic
decision experiment, Management science, 35, (3)

Sterman J. D. (1992) Teaching takes off, Flight simulators for management education, OR/MS Today,
October, 40-44

Vicente, K. J. and Rasmussen J. (1992) Ecological interface design: Theoretical foundations, JEEE
Transactions on systems, Man and Cybernetics, 22, (4).

15b